Blind Inversion Based On Bayesian Estimation

Seismic blind inversion is a kind of adaptive inversion method completely based on seismic data,it can be used to stratum in the no well or well data lack of elastic parameter prediction,and then to reservoir prediction and fluid identification which based on seismic data to provide data base.Seismic blind inversion mainly involves two aspects: first,the blind source separation of wavelet estimation problem;Second,the regularization of seismic inversion problems.First,As the new exploration closed wells data lack or exist big error,certainty seismic wavelet extraction methods cannot get effective implementation.In the absence of reliable log data a nd geological prior,it is particularly important to accurately estimate the seismic wavelet.Second,prestack seismic multi-parameter inversion blind post-stack seismic inversion is greater than not qualitative,quasi joint bayesian inversion method of building framework as well as the correlation between the objective function,that can improve the reliability of prestack seismic inversion method and the noise resistance.In the end,this article from the blind mixed phase wavelet extraction,blind post-stack inversion and prestack AVO inversion,elastic impedance inversion of blind inversion algorithm and the feasibility is verified.Blind prestack seismic inversion is the premise of high precision seismic wavelet estimation method,this paper uses the method of wavelet extraction is the differential evolution algorithm based on higher order statistics of seismic wavelet blind extraction method.First of all,based on the premise of seismic noise as gauss ian white noise;Then,according to the ideas of the higher-order cumulant blind construction of seismic wavelet extraction of target functional;Second,the search target functional based on differential evolution algorithm is global optimal solution,finally obtain high-precision mixed phase seismic wavelet.Differential evolution algorithm is global search efficient algorithm using real number coding,based on one-on-one crossover strategy,competition between individual species,the algorithm convergence speed,not easy to fall into local optimal solution,to improve the convergence speed and precision of the wavelet extraction.Through the theoretical model and actual data processing proves the effect of wavelet extraction,fully il ustrates the feasibility of this method.The wavelet be extracted by the seismic wavelet blind extraction method which represented above could be utilized for pre-stack seismic inversion.Under the Bayesian framework,the thesis studies the AVO inversion method,and the inversion method was demonstrated to be effective through the application of model testing and real data.The method can go well even there is no reliable data information,it can get the premise of a certain precision of inversion results,thus reduce the exploration cost.Finally,the thesis argues that if the two processes have the same or similar autocorrelation function,then the power spectrum of these two signals are the same or similar.Therefo re,the thesis proposed a novel inversion method named correlation-based inversion method which suggests characterizing the noise by a correlation analysis method,and L1 norm regularization was utilized to enhance the stability and sparsity of estimates.Theoretical model tests the stability of the novel method.In real data test,the result of this method is fit with the well log curve,and exhibit enormous potential in seismic exploration.And compared with the traditional method of calculating the elastic impedance,the method can shorten the inversion time,improve the inversion accuracy.